[Devel] Re: [PATCH 11/11] sysfs: user namespaces: fix bug with clone(CLONE_NEWUSER) with fairsched

[Serge E. Hallyn]

Quoting Eric W. Biederman (ebiederm at xmission.com):
> "Serge E. Hallyn" <serue at us.ibm.com> writes:
> 
> > Except the sysfs mount holds no refcount on the userns.  So as long as we
> > do the ida tagging as you suggested in your response to patch 6, there
> > should be no reference to the user_ns left in sysfs code.
> >
> > The extra reference in patch #9 is for a light ref on the network
> > namespace.  I'm still not sure that needs to be there, since if the
> > network namespace goes away, it will properly unregister its sysfs
> > mounts.  Eric, Benjamin, I really don't see any use for the hold_net()
> > from sysfs.  What is it doing?
> 
> Mostly just being a sanity check.  We can remove that if it easier.
> 
> Eric

In itself it seems an ok check (temporarily) to make sure that the
net_ns hook to unset the tag.netns at netns release is properly working,
but given that it's only checking for net_ns coding errors, and Tejun
wants a whole generic hooking infrastructure for net_ns to register with
so as to keep mention of net_ns out of sysfs code, yeah I think the sane
thing is to just remove it.

But the first thing (after Daniel is finished with the patch 5 fallout)
is to address the patch 6 comments about using ida.  At first I didn't
like it (seemed like too much bookkeeping) but I think it'll actually
work out very nicely.

-serge
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers